Guides
A MoE is five times faster and bigger in memory
A 26B mixture decodes 4 to 5 times faster than a dense model on a Mac and can take more RAM. Total parameters load, active ones compute.
10 minAI
Everything we publish on AI: the models we run on our own machines, what the tools send home, the benchmarks we measure ourselves and the claims we check against the code. Start with the investigations, then the guides.
Guides
Bonsai 27B or Gemma 4 31B on a Mac?
One is a fifth the size of the other, answers faster and holds more context; the hard questions decide it.
20 min
Investigations
How AI tools hide their trackers
A covert tracker in Claude Code, telemetry in others, and how to catch both.
8 min
Guides
How to stop a background agent failing silently
Ours was dead for three days while cron kept firing. A heartbeat, a wrapper and a watchdog raise the alarm within the hour.
7 min
Guides
A local agent that briefs you every morning
Hermes Agent set up to fold calendar, goals and weather into one private daily briefing.
8 min
Investigations
Graphify opened 358 sockets, all to LM Studio
We watched every connection over two passes: none left the desk, and the code graph cost zero tokens.
8 min
Investigations
OpenCode or OpenJarvis: which agent phones home?
We captured what each sends. One has a working off switch; the other, the agent we use daily, has none we could find.
7 min
Guides
A private assistant that never leaves the room
Hermes Agent and LM Studio on an M4 Pro, with voice notes on Telegram at 3 am and nothing sent to a cloud.
17 min
Guides
Three agents, one brand voice
One voice spec wired into OpenCode, Hermes and Pi through AGENTS.md, so their output matches.
12 min
Insights
Eight Mac minis can run a 671B model
exo pools Apple Silicon machines into one, so a cluster holds DeepSeek 671B that no single Mac can.
7 min